In davidthaler/PUMSutils: Computes Point Estimates and Standard Errors on the ACS PUMS Data

knitr::opts_chunk$set(echo = TRUE, fig.align = "center")
library(dplyr)
library(ggplot2)
library(PUMSutils)

This is a short introduction to plotting results for the ACS PUMS data with ggplot2 and PUMSutils. The main issue in plotting results drawn from the ACS PUMS data is that it is weighted sample. Each row represents a different number of cases in the population, indicated by the expansion weight (WGTP or PWGTP), and any correct calculation of a statistic has to use the expansion weights. There are basically 3 ways to proceed:

1. Some ggplot2 ojects take a weight aesthetic that can be set to WGTP or PWGTP.
2. PUMSutils has convenience functions for plotting. These set the weight aesthetic appropriately.
3. PUMSutils can be used to produce weighted statistics, which can then be plotted with geom_col.

We discuss each of these in turn. The data used in all plots is 1-year 2016 ACS data for Washington State.

Using the weight aesthetic in ggplot2

The first approach is to use ggplot, and set the weight aesthetic directly like this:

wa.house16$Tenure <- own.rent(wa.house16)
ggplot(wa.house16, aes(Tenure)) + geom_bar(aes(weight=WGTP))

Only some plot types accept a weight aesthetic. It can only be used for counts. And the feature is a little under-documented. For more details, see the ggplot docs.

PUMSutils convenience plotting functions

PUMSutils has convenience functions for plotting that cover some common cases. The functions take the most common labels as optional arguments. The acs.barchart function can produce barcharts of weighted count data. The plot above would be produced like this:

acs.barchart(wa.house16, 'Tenure', title='HH count by Tenure')

If you want to partition data on 2 variables, you can get a stacked barchart with acs.barchart by setting the stack argument. Below we chart employment status by gender.

wa.pop16$sex <- acs.recode(wa.pop16, 'SEX', data.dict16)
wa.pop16$Employment <- acs.recode(wa.pop16, 'ESR', data.dict16)
acs.barchart(wa.pop16, 'sex', stack='Employment', 
             title = 'Employment Status by Gender',
             xlab='Gender')

PUMSutils also includes acs.boxplot for producing weighted boxplots of the distribution of a numeric variable within categories. Here we plot the distribution of household income as a function of the number of workers in the household.

# ESR is employment status; 1,2,4,5 are employed
# match.count counts the number of people per household matching a condition
wa.house16$num.work <- match.count(wa.house16, wa.pop16, ESR %in% c(1,2,4,5))
wa.house16$num.work <- clip.column(wa.house16, 'num.work', 4)
wa.house16$inc.clip <- pmin(wa.house16$HINCP, 300000)
acs.boxplot(wa.house16[!is.na(wa.house16$HINCP),], 
            'num.work', 'inc.clip', 
            title='HH Income vs Number of Workers', 
            xlab='Number of Workers (max 4)', 
            ylab='HH Income($) (max $300K)')

Income variables have a lot of extreme outliers, so we clip income at 300,000.

Plot tabulated results with geom_col

If PUMSutils is used to produce weighted statistics, then those can be plotted using geom_col or similar from ggplot. Here we make a barchart of the median income of households as a function of the number of workers in the household.

med.inc <- group.median(wa.house16, 'num.work', 'HINCP', 
                        result.name = 'Median Income')
g <- ggplot(med.inc) + geom_col(aes(num.work, Median.Income))
g + ggtitle('Median HH Income vs. Number of Workers')

This is the only method that can plot weighted statistics other than counts.

davidthaler/PUMSutils documentation built on July 13, 2019, 9:58 a.m.